Conclusions First
- The control alone generates a woman floating in mid-air. Even without falling instructions, combining the base prompt with
full bodytends to produce aerial poses buildingadds an overhead city background with the strongest sense of falling from heightcliffconsistently reproduces a cliff-side coastal setting with a fall toward the seacloudscreates a floating composition among clouds — the highest sense of dreamy, weightless floatingalicereflects “Alice in Wonderland” with a circular hole fall composition and blue costume, but overwrites the white dressupside downbecomes a top-down photo of the subject lying on the ground face-up — not a falling expression at allslow motionemphasizes hair and water droplet movement in a studio-like floating pose — closer to a dance leap than a fallparachutegenerates a paragliding/parachute descent scene with harness and gear; outfit changes to sportswear
Experiment Design
| Item | Value |
|---|---|
| Model | z-image-turbo (6B, realism-focused distilled) |
| Steps | 8 |
| Sampler | euler |
| Scheduler | ddim_uniform |
| CFG | 1.0 |
| Size | 1024x1024 |
| Seeds | 3 fixed (shared across all conditions) |
| Per condition | 3 images (3 seeds) |
| Total | 8 conditions = 24 images |
Base Prompt
Each scene environment was varied per condition to observe scene reproduction.
A00: control
| seed 1 | seed 2 | seed 3 |
|---|---|---|
![]() | ![]() | ![]() |
Observation: All 3 images show a woman floating in the air. S1 uses a uniform gray background with a head-down falling pose. S2 and s3 show an overhead city composition with buildings and greenery below. A white dress flutters in the wind. Notably, falling/floating poses appear even without any scene keyword.
A01: building
| seed 1 | seed 2 | seed 3 |
|---|---|---|
![]() | ![]() | ![]() |
Observation: All 3 show a high-rise city background. S1 shows the subject grasping a building facade while falling, with skyscrapers below. S2 shows a face-up mid-air pose with arms spread above a dense high-rise skyline. S3 shows the subject leaning out from a rooftop railing. Compared to control, the building depiction is denser and taller, increasing the sense of altitude.
A02: cliff
| seed 1 | seed 2 | seed 3 |
|---|---|---|
![]() | ![]() | ![]() |
Observation: All 3 show a rocky coastal cliff. The subject consistently falls from the cliff edge toward the sea. S1 shows the moment of falling backward from the cliff; s2 shows the body tilted seaward while gripping the cliff edge; s3 shows a head-down fall. The background includes wave spray and detailed rock texture. Unlike building, this is a natural environment falling scene with consistent results.
A03: clouds
| seed 1 | seed 2 | seed 3 |
|---|---|---|
![]() | ![]() | ![]() |
Observation: All 3 show floating among or above clouds. S1 floats face-up surrounded by clouds in a blue sky — a dreamy, fantastical impression. S2 has a falling pose with hair flowing, ground visible through cloud gaps. S3 floats above clouds, the ground dimly visible below, with a white puff-sleeve dress spreading in the breeze. The composition is more “floating in the sky” than “falling,” with a quiet, fantastical atmosphere.
Lab Director comment: This clouds floating thing… what is this. It’s not falling — it’s more like dissolving into the sky. I love how the white dress and the clouds seem to merge together.
A04: alice
| seed 1 | seed 2 | seed 3 |
|---|---|---|
![]() | ![]() | ![]() |
Observation: All 3 generated a composition falling into a circular hole in the ground, surrounded by grass and soil — reminiscent of the rabbit hole in Alice in Wonderland. The costume changed to a blue dress regardless of the base prompt specification. S1 shows the subject spread out near the hole entrance; s2 shows a feet-up fall into the hole; s3 shows the subject spread out midway through a tunnel. A completely unique scene compared to all other conditions.
A05: upside down
| seed 1 | seed 2 | seed 3 |
|---|---|---|
![]() | ![]() | ![]() |
Observation: All 3 generated a top-down photo of the subject lying face-up on asphalt, not a falling scene. Hair spreads on the ground, limbs extended. No floating sensation — a shadow falls on the ground. “Upside down” means inverted, but the model interpreted it as a “vertically flipped overhead composition,” which does not function as a falling expression.
A06: slow motion
| seed 1 | seed 2 | seed 3 |
|---|---|---|
![]() | ![]() | ![]() |
Observation: All 3 show a dynamic mid-jump moment in an indoor or near-ground outdoor setting. S1 shows a studio-like space with a backward arching fall pose, hair flowing dramatically, water-drop-like particles at the feet. S2 has a gray wall background with a dynamic upward-looking pose with flowing hair. S3 shows an outdoor jump through water spray. All 3 are near-ground dynamic snapshots, not a high-altitude fall. Emphasized hair and clothing movement is distinctive.
A07: parachute
| seed 1 | seed 2 | seed 3 |
|---|---|---|
![]() | ![]() | ![]() |
Observation: All 3 show a paragliding or parachute descent scene. Harness, rope, and canopy are depicted. S1 shows a blue jumpsuit with helmet; s2 shows white clothing with harness gripping paraglider ropes; s3 shows a red and blue parachute canopy fully open while descending. Mountains and settlements visible in the background with a sense of altitude. The white dress is replaced with sportswear or a jumpsuit — outfit is overwritten.
Lab Director Comment
building and cliff have the strongest falling sensation; clouds is in the fantasy/floating camp; alice is its own unique category — these roles are cleanly separated. The fact that upside down becomes an overhead lying-on-ground photo is a complete trap, so I’d avoid it for falling expressions.
Summary
| Condition | Falling feel | Scene characteristics | Outfit maintained |
|---|---|---|---|
| control | Medium | Urban aerial, gray background | Maintained |
| building | High | Dense high-rise city, strong altitude | Maintained |
| cliff | High | Coastal cliff fall toward sea | Maintained |
| clouds | Medium | Floating among clouds, fantastical | Maintained |
| alice | Medium | Fall into rabbit hole | Changes to blue |
| upside down | None | Overhead photo of person lying on ground | Maintained |
| slow motion | Low | Near-ground leap, movement emphasized | Maintained |
| parachute | Medium | Paraglider descent | Changes to sportswear |
For falling scenes, building and cliff consistently generate the most altitude-intense falling compositions. clouds has more floating than falling sensation, suitable for fantastical imagery. alice generates a unique falling-into-a-hole composition but overwrites the outfit. upside down does not function as a falling expression — use with caution. slow motion becomes a dance-like leap rather than a fall. parachute generates an equipped descent scene and changes the outfit.
![[Experiment] Scene Setting Prompts for Falling | 8 Conditions, 24 Images](/tips/falling-scene-test/a03_clouds_s2_0_0000_7295072554507705269.webp)
























![[Experiment] Prompts for Mid-Fall Expressions | 13 Conditions, 39 Images](/tips/falling-expression-test/a02_falling_sky_s2_0_0000_7295072554507705269.webp)